static void
intel_bufferobj_copy_subdata(struct gl_context *ctx,
struct gl_buffer_object *src,
struct gl_buffer_object *dst,
GLintptr read_offset, GLintptr write_offset,
GLsizeiptr size)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_src = intel_buffer_object(src);
struct intel_buffer_object *intel_dst = intel_buffer_object(dst);
drm_intel_bo *src_bo, *dst_bo;
GLuint src_offset;
if (size == 0)
return;
/* If we're in system memory, just map and memcpy. */
if (intel_src->sys_buffer || intel_dst->sys_buffer || intel->gen >= 6) {
/* The same buffer may be used, but note that regions copied may
* not overlap.
*/
if (src == dst) {
char *ptr = intel_bufferobj_map_range(ctx, 0, dst->Size,
GL_MAP_READ_BIT, dst);
memmove(ptr + write_offset, ptr + read_offset, size);
intel_bufferobj_unmap(ctx, dst);
} else {
const char *src_ptr;
char *dst_ptr;
src_ptr = intel_bufferobj_map_range(ctx, 0, src->Size,
GL_MAP_READ_BIT, src);
dst_ptr = intel_bufferobj_map_range(ctx, 0, dst->Size,
GL_MAP_WRITE_BIT, dst);
memcpy(dst_ptr + write_offset, src_ptr + read_offset, size);
intel_bufferobj_unmap(ctx, src);
intel_bufferobj_unmap(ctx, dst);
}
return;
}
/* Otherwise, we have real BOs, so blit them. */
dst_bo = intel_bufferobj_buffer(intel, intel_dst, INTEL_WRITE_PART);
src_bo = intel_bufferobj_source(intel, intel_src, 64, &src_offset);
intel_emit_linear_blit(intel,
dst_bo, write_offset,
src_bo, read_offset + src_offset, size);
/* Since we've emitted some blits to buffers that will (likely) be used
* in rendering operations in other cache domains in this batch, emit a
* flush. Once again, we wish for a domain tracker in libdrm to cover
* usage inside of a batchbuffer.
*/
intel_batchbuffer_emit_mi_flush(intel);
}
static GLenum
intel_buffer_object_purgeable(struct gl_context * ctx,
struct gl_buffer_object *obj,
GLenum option)
{
struct intel_buffer_object *intel_obj = intel_buffer_object (obj);
if (intel_obj->buffer != NULL)
return intel_buffer_purgeable(intel_obj->buffer);
if (option == GL_RELEASED_APPLE) {
if (intel_obj->sys_buffer != NULL) {
free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
}
return GL_RELEASED_APPLE;
} else {
/* XXX Create the buffer and madvise(MADV_DONTNEED)? */
struct intel_context *intel = intel_context(ctx);
drm_intel_bo *bo = intel_bufferobj_buffer(intel, intel_obj, INTEL_READ);
return intel_buffer_purgeable(bo);
}
}
void brw_draw_init( struct brw_context *brw )
{
GLcontext *ctx = &brw->intel.ctx;
struct vbo_context *vbo = vbo_context(ctx);
GLuint i;
/* Register our drawing function:
*/
vbo->draw_prims = brw_draw_prims;
brw->vb.upload.size = BRW_UPLOAD_INIT_SIZE;
for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
/* NOTE: These are set to no-backing-store.
*/
bmBufferSetInvalidateCB(&brw->intel,
intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
brw_invalidate_vbo_cb,
&brw->intel,
GL_TRUE);
}
ctx->Driver.BufferData( ctx,
GL_ARRAY_BUFFER_ARB,
BRW_UPLOAD_INIT_SIZE,
NULL,
GL_DYNAMIC_DRAW_ARB,
brw->vb.upload.vbo[0] );
}
/**
* The BufferData() driver hook.
*
* Implements glBufferData(), which recreates a buffer object's data store
* and populates it with the given data, if present.
*
* Any data that was previously stored in the buffer object is lost.
*
* \return true for success, false if out of memory
*/
static GLboolean
intel_bufferobj_data(struct gl_context * ctx,
GLenum target,
GLsizeiptrARB size,
const GLvoid * data,
GLenum usage, struct gl_buffer_object *obj)
{
struct brw_context *brw = brw_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
/* Part of the ABI, but this function doesn't use it.
*/
(void) target;
intel_obj->Base.Size = size;
intel_obj->Base.Usage = usage;
assert(!obj->Pointer); /* Mesa should have unmapped it */
if (intel_obj->buffer != NULL)
release_buffer(intel_obj);
if (size != 0) {
intel_bufferobj_alloc_buffer(brw, intel_obj);
if (!intel_obj->buffer)
return false;
if (data != NULL)
drm_intel_bo_subdata(intel_obj->buffer, 0, size, data);
}
return true;
}
/**
* Deallocate/free a vertex/pixel buffer object.
* Called via glDeleteBuffersARB().
*/
static void
intel_bufferobj_free(GLcontext * ctx, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
/* Buffer objects are automatically unmapped when deleting according
* to the spec, but Mesa doesn't do UnmapBuffer for us at context destroy
* (though it does if you call glDeleteBuffers)
*/
if (obj->Pointer)
intel_bufferobj_unmap(ctx, 0, obj);
_mesa_free(intel_obj->sys_buffer);
if (intel_obj->region) {
intel_bufferobj_release_region(intel, intel_obj);
}
else if (intel_obj->buffer) {
dri_bo_unreference(intel_obj->buffer);
}
_mesa_free(intel_obj);
}
/* Ideally we'd use a BO to avoid taking up cache space for the temporary
* data, but FlushMappedBufferRange may be followed by further writes to
* the pointer, so we would have to re-map after emitting our blit, which
* would defeat the point.
*/
static void
intel_bufferobj_flush_mapped_range(GLcontext *ctx, GLenum target,
GLintptr offset, GLsizeiptr length,
struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
drm_intel_bo *temp_bo;
/* Unless we're in the range map using a temporary system buffer,
* there's no work to do.
*/
if (intel_obj->range_map_buffer == NULL)
return;
temp_bo = drm_intel_bo_alloc(intel->bufmgr, "range map flush", length, 64);
drm_intel_bo_subdata(temp_bo, 0, length, intel_obj->range_map_buffer);
intel_emit_linear_blit(intel,
intel_obj->buffer, obj->Offset + offset,
temp_bo, 0,
length);
drm_intel_bo_unreference(temp_bo);
}
static GLenum
intel_buffer_object_unpurgeable(GLcontext * ctx,
struct gl_buffer_object *obj,
GLenum option)
{
return intel_buffer_unpurgeable (ctx, intel_buffer_object (obj)->buffer, option);
}
/**
* Replace data in a subrange of buffer object. If the data range
* specified by size + offset extends beyond the end of the buffer or
* if data is NULL, no copy is performed.
* Called via glBufferSubDataARB().
*/
static void
intel_bufferobj_subdata(struct gl_context * ctx,
GLintptrARB offset,
GLsizeiptrARB size,
const GLvoid * data, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
bool busy;
if (size == 0)
return;
assert(intel_obj);
/* If we have a single copy in system memory, update that */
if (intel_obj->sys_buffer) {
if (intel_obj->source)
release_buffer(intel_obj);
if (intel_obj->buffer == NULL) {
memcpy((char *)intel_obj->sys_buffer + offset, data, size);
return;
}
free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
}
/* Otherwise we need to update the copy in video memory. */
busy =
drm_intel_bo_busy(intel_obj->buffer) ||
drm_intel_bo_references(intel->batch.bo, intel_obj->buffer);
if (busy) {
if (size == intel_obj->Base.Size) {
/* Replace the current busy bo with fresh data. */
drm_intel_bo_unreference(intel_obj->buffer);
intel_bufferobj_alloc_buffer(intel, intel_obj);
drm_intel_bo_subdata(intel_obj->buffer, 0, size, data);
} else {
perf_debug("Using a blit copy to avoid stalling on %ldb "
"glBufferSubData() to a busy buffer object.\n",
(long)size);
drm_intel_bo *temp_bo =
drm_intel_bo_alloc(intel->bufmgr, "subdata temp", size, 64);
drm_intel_bo_subdata(temp_bo, 0, size, data);
intel_emit_linear_blit(intel,
intel_obj->buffer, offset,
temp_bo, 0,
size);
drm_intel_bo_unreference(temp_bo);
}
} else {
drm_intel_bo_subdata(intel_obj->buffer, offset, size, data);
}
}
static void
gen8_upload_3dstate_so_buffers(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
struct brw_transform_feedback_object *brw_obj =
(struct brw_transform_feedback_object *) xfb_obj;
/* Set up the up to 4 output buffers. These are the ranges defined in the
* gl_transform_feedback_object.
*/
for (int i = 0; i < 4; i++) {
struct intel_buffer_object *bufferobj =
intel_buffer_object(xfb_obj->Buffers[i]);
if (!bufferobj) {
BEGIN_BATCH(8);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (8 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
continue;
}
uint32_t start = xfb_obj->Offset[i];
assert(start % 4 == 0);
uint32_t end = ALIGN(start + xfb_obj->Size[i], 4);
drm_intel_bo *bo =
intel_bufferobj_buffer(brw, bufferobj, start, end - start);
assert(end <= bo->size);
perf_debug("Missing MOCS setup for 3DSTATE_SO_BUFFER.");
BEGIN_BATCH(8);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (8 - 2));
OUT_BATCH(GEN8_SO_BUFFER_ENABLE | (i << SO_BUFFER_INDEX_SHIFT) |
GEN8_SO_BUFFER_OFFSET_WRITE_ENABLE |
GEN8_SO_BUFFER_OFFSET_ADDRESS_ENABLE |
(BDW_MOCS_WB << 22));
OUT_RELOC64(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, start);
OUT_BATCH(xfb_obj->Size[i] / 4 - 1);
OUT_RELOC64(brw_obj->offset_bo,
I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
i * sizeof(uint32_t));
if (brw_obj->zero_offsets)
OUT_BATCH(0); /* Zero out the offset and write that to offset_bo */
else
OUT_BATCH(0xFFFFFFFF); /* Use offset_bo as the "Stream Offset." */
ADVANCE_BATCH();
}
brw_obj->zero_offsets = false;
}
static bool
intel_set_texture_storage_for_buffer_object(struct gl_context *ctx,
struct gl_texture_object *tex_obj,
struct gl_buffer_object *buffer_obj,
uint32_t buffer_offset,
uint32_t row_stride,
bool read_only)
{
struct brw_context *brw = brw_context(ctx);
struct intel_texture_object *intel_texobj = intel_texture_object(tex_obj);
struct gl_texture_image *image = tex_obj->Image[0][0];
struct intel_texture_image *intel_image = intel_texture_image(image);
struct intel_buffer_object *intel_buffer_obj = intel_buffer_object(buffer_obj);
if (!read_only) {
/* Renderbuffers have the restriction that the buffer offset and
* surface pitch must be a multiple of the element size. If it's
* not, we have to fail and fall back to software.
*/
int cpp = _mesa_get_format_bytes(image->TexFormat);
if (buffer_offset % cpp || row_stride % cpp) {
perf_debug("Bad PBO alignment; fallback to CPU mapping\n");
return false;
}
if (!brw->format_supported_as_render_target[image->TexFormat]) {
perf_debug("Non-renderable PBO format; fallback to CPU mapping\n");
return false;
}
}
assert(intel_texobj->mt == NULL);
drm_intel_bo *bo = intel_bufferobj_buffer(brw, intel_buffer_obj,
buffer_offset,
row_stride * image->Height);
intel_texobj->mt =
intel_miptree_create_for_bo(brw, bo,
image->TexFormat,
buffer_offset,
image->Width, image->Height, image->Depth,
row_stride,
0);
if (!intel_texobj->mt)
return false;
if (!_swrast_init_texture_image(image))
return false;
intel_miptree_reference(&intel_image->mt, intel_texobj->mt);
/* The miptree is in a validated state, so no need to check later. */
intel_texobj->needs_validate = false;
intel_texobj->validated_first_level = 0;
intel_texobj->validated_last_level = 0;
intel_texobj->_Format = intel_texobj->mt->format;
return true;
}
static GLenum
intel_buffer_object_unpurgeable(struct gl_context * ctx,
struct gl_buffer_object *obj,
GLenum option)
{
(void) ctx;
(void) option;
return intel_buffer_unpurgeable(intel_buffer_object(obj)->buffer);
}
/* XXX: Do this for TexSubImage also:
*/
static GLboolean
try_pbo_upload(struct intel_context *intel,
struct intel_texture_image *intelImage,
const struct gl_pixelstore_attrib *unpack,
GLint internalFormat,
GLint width, GLint height,
GLenum format, GLenum type, const void *pixels)
{
struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
GLuint src_offset, src_stride;
GLuint dst_offset, dst_stride;
if (!pbo ||
intel->ctx._ImageTransferState ||
unpack->SkipPixels || unpack->SkipRows) {
_mesa_printf("%s: failure 1\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) pixels;
if (unpack->RowLength > 0)
src_stride = unpack->RowLength;
else
src_stride = width;
dst_offset = intel_miptree_image_offset(intelImage->mt,
intelImage->face,
intelImage->level);
dst_stride = intelImage->mt->pitch;
intelFlush(&intel->ctx);
LOCK_HARDWARE(intel);
{
struct _DriBufferObject *src_buffer =
intel_bufferobj_buffer(intel, pbo, INTEL_READ);
struct _DriBufferObject *dst_buffer =
intel_region_buffer(intel->intelScreen, intelImage->mt->region,
INTEL_WRITE_FULL);
intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_stride, src_buffer, src_offset,
dst_stride, dst_buffer, dst_offset,
0, 0, 0, 0, width, height,
GL_COPY);
intel_batchbuffer_flush(intel->batch);
}
UNLOCK_HARDWARE(intel);
return GL_TRUE;
}
/**
* Called via glMapBufferARB().
*/
static void *
intel_bufferobj_map(struct gl_context * ctx,
GLenum target,
GLenum access, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
GLboolean read_only = (access == GL_READ_ONLY_ARB);
GLboolean write_only = (access == GL_WRITE_ONLY_ARB);
assert(intel_obj);
if (intel_obj->sys_buffer) {
if (!read_only && intel_obj->source) {
release_buffer(intel_obj);
}
if (!intel_obj->buffer || intel_obj->source) {
obj->Pointer = intel_obj->sys_buffer;
obj->Length = obj->Size;
obj->Offset = 0;
return obj->Pointer;
}
free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
}
/* Flush any existing batchbuffer that might reference this data. */
if (drm_intel_bo_references(intel->batch.bo, intel_obj->buffer))
intel_flush(ctx);
if (intel_obj->region)
intel_bufferobj_cow(intel, intel_obj);
if (intel_obj->buffer == NULL) {
obj->Pointer = NULL;
return NULL;
}
if (write_only) {
drm_intel_gem_bo_map_gtt(intel_obj->buffer);
intel_obj->mapped_gtt = GL_TRUE;
} else {
drm_intel_bo_map(intel_obj->buffer, !read_only);
intel_obj->mapped_gtt = GL_FALSE;
}
obj->Pointer = intel_obj->buffer->virtual;
obj->Length = obj->Size;
obj->Offset = 0;
return obj->Pointer;
}
static void
upload_3dstate_so_buffers(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_VERTEX_PROGRAM */
const struct gl_shader_program *vs_prog =
ctx->Shader.CurrentVertexProgram;
const struct gl_transform_feedback_info *linked_xfb_info =
&vs_prog->LinkedTransformFeedback;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
int i;
/* Set up the up to 4 output buffers. These are the ranges defined in the
* gl_transform_feedback_object.
*/
for (i = 0; i < 4; i++) {
struct intel_buffer_object *bufferobj =
intel_buffer_object(xfb_obj->Buffers[i]);
drm_intel_bo *bo;
uint32_t start, end;
uint32_t stride;
if (!xfb_obj->Buffers[i]) {
/* The pitch of 0 in this command indicates that the buffer is
* unbound and won't be written to.
*/
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (4 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
continue;
}
bo = intel_bufferobj_buffer(brw, bufferobj, INTEL_WRITE_PART);
stride = linked_xfb_info->BufferStride[i] * 4;
start = xfb_obj->Offset[i];
assert(start % 4 == 0);
end = ALIGN(start + xfb_obj->Size[i], 4);
assert(end <= bo->size);
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (4 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT) | stride);
OUT_RELOC(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, start);
OUT_RELOC(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, end);
ADVANCE_BATCH();
}
}
/**
* Called via glUnmapBuffer().
*/
static GLboolean
intel_bufferobj_unmap(GLcontext * ctx,
GLenum target, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
assert(obj->Pointer);
if (intel_obj->sys_buffer != NULL) {
/* always keep the mapping around. */
} else if (intel_obj->range_map_buffer != NULL) {
/* Since we've emitted some blits to buffers that will (likely) be used
* in rendering operations in other cache domains in this batch, emit a
* flush. Once again, we wish for a domain tracker in libdrm to cover
* usage inside of a batchbuffer.
*/
intel_batchbuffer_emit_mi_flush(intel->batch);
free(intel_obj->range_map_buffer);
intel_obj->range_map_buffer = NULL;
} else if (intel_obj->range_map_bo != NULL) {
if (intel_obj->mapped_gtt) {
drm_intel_gem_bo_unmap_gtt(intel_obj->range_map_bo);
} else {
drm_intel_bo_unmap(intel_obj->range_map_bo);
}
intel_emit_linear_blit(intel,
intel_obj->buffer, obj->Offset,
intel_obj->range_map_bo, 0,
obj->Length);
/* Since we've emitted some blits to buffers that will (likely) be used
* in rendering operations in other cache domains in this batch, emit a
* flush. Once again, we wish for a domain tracker in libdrm to cover
* usage inside of a batchbuffer.
*/
intel_batchbuffer_emit_mi_flush(intel->batch);
drm_intel_bo_unreference(intel_obj->range_map_bo);
intel_obj->range_map_bo = NULL;
} else if (intel_obj->buffer != NULL) {
if (intel_obj->mapped_gtt) {
drm_intel_gem_bo_unmap_gtt(intel_obj->buffer);
} else {
drm_intel_bo_unmap(intel_obj->buffer);
}
}
obj->Pointer = NULL;
obj->Offset = 0;
obj->Length = 0;
return GL_TRUE;
}
/**
* Allocate space for and store data in a buffer object. Any data that was
* previously stored in the buffer object is lost. If data is NULL,
* memory will be allocated, but no copy will occur.
* Called via ctx->Driver.BufferData().
* \return true for success, false if out of memory
*/
static GLboolean
intel_bufferobj_data(struct gl_context * ctx,
GLenum target,
GLsizeiptrARB size,
const GLvoid * data,
GLenum usage, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
/* Part of the ABI, but this function doesn't use it.
*/
#ifndef I915
(void) target;
#endif
intel_obj->Base.Size = size;
intel_obj->Base.Usage = usage;
assert(!obj->Pointer); /* Mesa should have unmapped it */
if (intel_obj->buffer != NULL)
release_buffer(intel_obj);
free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
if (size != 0) {
if (usage == GL_DYNAMIC_DRAW
#ifdef I915
/* On pre-965, stick VBOs in system memory, as we're always doing
* swtnl with their contents anyway.
*/
|| target == GL_ARRAY_BUFFER || target == GL_ELEMENT_ARRAY_BUFFER
#endif
)
{
intel_obj->sys_buffer = malloc(size);
if (intel_obj->sys_buffer != NULL) {
if (data != NULL)
memcpy(intel_obj->sys_buffer, data, size);
return true;
}
}
intel_bufferobj_alloc_buffer(intel, intel_obj);
if (!intel_obj->buffer)
return false;
if (data != NULL)
drm_intel_bo_subdata(intel_obj->buffer, 0, size, data);
}
return true;
}
int brw_prepare_indices( struct brw_context *brw,
const struct _mesa_index_buffer *index_buffer,
dri_bo **bo_return,
GLuint *offset_return)
{
GLcontext *ctx = &brw->intel.ctx;
struct intel_context *intel = &brw->intel;
GLuint ib_size = get_size(index_buffer->type) * index_buffer->count;
dri_bo *bo;
struct gl_buffer_object *bufferobj = index_buffer->obj;
GLuint offset = (GLuint)index_buffer->ptr;
int ret;
/* Turn into a proper VBO:
*/
if (!bufferobj->Name) {
/* Get new bufferobj, offset:
*/
get_space(brw, ib_size, &bo, &offset);
/* Straight upload
*/
dri_bo_subdata(bo, offset, ib_size, index_buffer->ptr);
} else {
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
*/
if ((get_size(index_buffer->type) - 1) & offset) {
GLubyte *map = ctx->Driver.MapBuffer(ctx,
GL_ELEMENT_ARRAY_BUFFER_ARB,
GL_DYNAMIC_DRAW_ARB,
bufferobj);
map += offset;
get_space(brw, ib_size, &bo, &offset);
dri_bo_subdata(bo, offset, ib_size, map);
ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
} else {
bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj),
INTEL_READ);
dri_bo_reference(bo);
}
}
*bo_return = bo;
*offset_return = offset;
ret = dri_bufmgr_check_aperture_space(bo);
return ret;
}
/**
* Called via glGetBufferSubDataARB().
*/
static void intel_bufferobj_get_subdata( GLcontext *ctx,
GLenum target,
GLintptrARB offset,
GLsizeiptrARB size,
GLvoid * data,
struct gl_buffer_object * obj )
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
bmBufferGetSubData(intel->bm, intel_obj->buffer, offset, size, data);
}
/**
* Deallocate/free a vertex/pixel buffer object.
* Called via glDeleteBuffersARB().
*/
static void intel_bufferobj_free( GLcontext *ctx,
struct gl_buffer_object *obj )
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
if (intel_obj->buffer)
bmDeleteBuffers( intel->bm, 1, &intel_obj->buffer );
_mesa_free(intel_obj);
}
/**
* Called via glMapBufferARB().
*/
static GLboolean intel_bufferobj_unmap( GLcontext *ctx,
GLenum target,
struct gl_buffer_object *obj )
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
assert(obj->Pointer);
bmUnmapBuffer(intel->bm, intel_obj->buffer);
obj->Pointer = NULL;
return GL_TRUE;
}
/**
* Called via glMapBufferARB().
*/
static void *intel_bufferobj_map( GLcontext *ctx,
GLenum target,
GLenum access,
struct gl_buffer_object *obj )
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
/* XXX: Translate access to flags arg below:
*/
assert(intel_obj);
obj->Pointer = bmMapBuffer(intel->bm, intel_obj->buffer, 0);
return obj->Pointer;
}
/**
* Allocate space for and store data in a buffer object. Any data that was
* previously stored in the buffer object is lost. If data is NULL,
* memory will be allocated, but no copy will occur.
* Called via ctx->Driver.BufferData().
* \return GL_TRUE for success, GL_FALSE if out of memory
*/
static GLboolean
intel_bufferobj_data(GLcontext * ctx,
GLenum target,
GLsizeiptrARB size,
const GLvoid * data,
GLenum usage, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
intel_obj->Base.Size = size;
intel_obj->Base.Usage = usage;
assert(!obj->Pointer); /* Mesa should have unmapped it */
if (intel_obj->region)
intel_bufferobj_release_region(intel, intel_obj);
if (intel_obj->buffer != NULL) {
dri_bo_unreference(intel_obj->buffer);
intel_obj->buffer = NULL;
}
_mesa_free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
if (size != 0) {
#ifdef I915
/* On pre-965, stick VBOs in system memory, as we're always doing swtnl
* with their contents anyway.
*/
if (target == GL_ARRAY_BUFFER || target == GL_ELEMENT_ARRAY_BUFFER) {
intel_obj->sys_buffer = _mesa_malloc(size);
if (intel_obj->sys_buffer != NULL) {
if (data != NULL)
memcpy(intel_obj->sys_buffer, data, size);
return GL_TRUE;
}
}
#endif
intel_bufferobj_alloc_buffer(intel, intel_obj);
if (!intel_obj->buffer)
return GL_FALSE;
if (data != NULL)
dri_bo_subdata(intel_obj->buffer, 0, size, data);
}
return GL_TRUE;
}
/**
* Called via glGetBufferSubDataARB().
*/
static void
intel_bufferobj_get_subdata(GLcontext * ctx,
GLenum target,
GLintptrARB offset,
GLsizeiptrARB size,
GLvoid * data, struct gl_buffer_object *obj)
{
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
if (intel_obj->sys_buffer)
memcpy(data, (char *)intel_obj->sys_buffer + offset, size);
else
dri_bo_get_subdata(intel_obj->buffer, offset, size, data);
}
/**
* The DeleteBuffer() driver hook.
*
* Deletes a single OpenGL buffer object. Used by glDeleteBuffers().
*/
static void
brw_delete_buffer(struct gl_context * ctx, struct gl_buffer_object *obj)
{
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
assert(intel_obj);
/* Buffer objects are automatically unmapped when deleting according
* to the spec, but Mesa doesn't do UnmapBuffer for us at context destroy
* (though it does if you call glDeleteBuffers)
*/
_mesa_buffer_unmap_all_mappings(ctx, obj);
drm_intel_bo_unreference(intel_obj->buffer);
free(intel_obj);
}
static GLenum
intel_buffer_object_purgeable(struct gl_context * ctx,
struct gl_buffer_object *obj,
GLenum option)
{
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
if (intel_obj->buffer != NULL)
return intel_buffer_purgeable(intel_obj->buffer);
if (option == GL_RELEASED_APPLE) {
return GL_RELEASED_APPLE;
} else {
/* XXX Create the buffer and madvise(MADV_DONTNEED)? */
return intel_buffer_purgeable(intel_obj->buffer);
}
}
/**
* Called via glGetBufferSubDataARB().
*/
static void
intel_bufferobj_get_subdata(struct gl_context * ctx,
GLintptrARB offset,
GLsizeiptrARB size,
GLvoid * data, struct gl_buffer_object *obj)
{
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
struct intel_context *intel = intel_context(ctx);
assert(intel_obj);
if (intel_obj->sys_buffer)
memcpy(data, (char *)intel_obj->sys_buffer + offset, size);
else {
if (drm_intel_bo_references(intel->batch.bo, intel_obj->buffer)) {
intel_batchbuffer_flush(intel);
}
drm_intel_bo_get_subdata(intel_obj->buffer, offset, size, data);
}
}
/**
* Allocate space for and store data in a buffer object. Any data that was
* previously stored in the buffer object is lost. If data is NULL,
* memory will be allocated, but no copy will occur.
* Called via glBufferDataARB().
*/
static void intel_bufferobj_data( GLcontext *ctx,
GLenum target,
GLsizeiptrARB size,
const GLvoid *data,
GLenum usage,
struct gl_buffer_object *obj )
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
/* XXX: do something useful with 'usage' (eg. populate flags
* argument below)
*/
assert(intel_obj);
obj->Size = size;
obj->Usage = usage;
bmBufferData(intel->bm, intel_obj->buffer, size, data, 0);
}
static GLboolean
try_pbo_zcopy(struct intel_context *intel,
struct intel_texture_image *intelImage,
const struct gl_pixelstore_attrib *unpack,
GLint internalFormat,
GLint width, GLint height,
GLenum format, GLenum type, const void *pixels)
{
struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
GLuint src_offset, src_stride;
GLuint dst_offset, dst_stride;
if (!pbo ||
intel->ctx._ImageTransferState ||
unpack->SkipPixels || unpack->SkipRows) {
_mesa_printf("%s: failure 1\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) pixels;
if (unpack->RowLength > 0)
src_stride = unpack->RowLength;
else
src_stride = width;
dst_offset = intel_miptree_image_offset(intelImage->mt,
intelImage->face,
intelImage->level);
dst_stride = intelImage->mt->pitch;
if (src_stride != dst_stride || dst_offset != 0 || src_offset != 0) {
_mesa_printf("%s: failure 2\n", __FUNCTION__);
return GL_FALSE;
}
intel_region_attach_pbo(intel->intelScreen, intelImage->mt->region, pbo);
return GL_TRUE;
}
/**
* Replace data in a subrange of buffer object. If the data range
* specified by size + offset extends beyond the end of the buffer or
* if data is NULL, no copy is performed.
* Called via glBufferSubDataARB().
*/
static void
intel_bufferobj_subdata(GLcontext * ctx,
GLenum target,
GLintptrARB offset,
GLsizeiptrARB size,
const GLvoid * data, struct gl_buffer_object *obj)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
if (size == 0)
return;
assert(intel_obj);
if (intel_obj->region)
intel_bufferobj_cow(intel, intel_obj);
if (intel_obj->sys_buffer)
memcpy((char *)intel_obj->sys_buffer + offset, data, size);
else {
/* Flush any existing batchbuffer that might reference this data. */
if (drm_intel_bo_busy(intel_obj->buffer) ||
drm_intel_bo_references(intel->batch->buf, intel_obj->buffer)) {
drm_intel_bo *temp_bo;
temp_bo = drm_intel_bo_alloc(intel->bufmgr, "subdata temp", size, 64);
drm_intel_bo_subdata(temp_bo, 0, size, data);
intel_emit_linear_blit(intel,
intel_obj->buffer, offset,
temp_bo, 0,
size);
drm_intel_bo_unreference(temp_bo);
} else {
drm_intel_bo_subdata(intel_obj->buffer, offset, size, data);
}
}
}
/**
* The BufferSubData() driver hook.
*
* Implements glBufferSubData(), which replaces a portion of the data in a
* buffer object.
*
* If the data range specified by (size + offset) extends beyond the end of
* the buffer or if data is NULL, no copy is performed.
*/
static void
brw_buffer_subdata(struct gl_context *ctx,
GLintptrARB offset,
GLsizeiptrARB size,
const GLvoid *data,
struct gl_buffer_object *obj)
{
struct brw_context *brw = brw_context(ctx);
struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
bool busy;
if (size == 0)
return;
assert(intel_obj);
/* See if we can unsynchronized write the data into the user's BO. This
* avoids GPU stalls in unfortunately common user patterns (uploading
* sequentially into a BO, with draw calls in between each upload).
*
* Once we've hit this path, we mark this GL BO as preferring stalling to
* blits, so that we can hopefully hit this path again in the future
* (otherwise, an app that might occasionally stall but mostly not will end
* up with blitting all the time, at the cost of bandwidth)
*/
if (brw->has_llc) {
if (offset + size <= intel_obj->gpu_active_start ||
intel_obj->gpu_active_end <= offset) {
drm_intel_gem_bo_map_unsynchronized(intel_obj->buffer);
memcpy(intel_obj->buffer->virtual + offset, data, size);
drm_intel_bo_unmap(intel_obj->buffer);
if (intel_obj->gpu_active_end > intel_obj->gpu_active_start)
intel_obj->prefer_stall_to_blit = true;
return;
}
}
